Ghosh and Sowdhamini BMC Genomics (2017) 18:658 DOI 10.1186/s12864-017-4045-3

RESEARCH ARTICLE Open Access Bioinformatics comparisons of RNA-binding proteins of pathogenic and non-pathogenic Escherichia coli strains reveal novel virulence factors Pritha Ghosh and Ramanathan Sowdhamini*

Abstract Background: Pathogenic bacteria have evolved various strategies to counteract host defences. They are also exposed to environments that are undergoing constant changes. Hence, in order to survive, bacteria must adapt themselves to the changing environmental conditions by performing regulations at the transcriptional and/or post-transcriptional levels. Roles of RNA-binding proteins (RBPs) as virulence factors have been very well studied. Here, we have used a sequence search-based method to compare and contrast the proteomes of 16 pathogenic and three non-pathogenic E. coli strains as well as to obtain a global picture of the RBP landscape (RBPome) in E. coli. Results: Our results show that there are no significant differences in the percentage of RBPs encoded by the pathogenic and the non-pathogenic E. coli strains. The differences in the types of Pfam domains as well as Pfam RNA-binding domains, encoded by these two classes of E. coli strains, are also insignificant. The complete and distinct RBPome of E. coli has been established by studying all known E. coli strains till date. We have also identified RBPs that are exclusive to pathogenic strains, and most of them can be exploited as drug targets since they appear to be non-homologous to their human host proteins. Many of these pathogen-specific proteins were uncharacterised and their identities could be resolved on the basis of sequence homology searches with known proteins. Detailed structural modelling, molecular dynamics simulations and sequence comparisons have been pursued for selected examples to understand differences in stability and RNA-binding. Conclusions: The approach used in this paper to cross-compare proteomes of pathogenic and non-pathogenic strains may also be extended to other bacterial or even eukaryotic proteomes to understand interesting differences in their RBPomes. The pathogen-specific RBPs reported in this study, may also be taken up further for clinical trials and/or experimental validations. Keywords: Escherichia coli, RNA-binding proteins, Genome-wide survey, Pathogen, Virulence, Ribonuclease PH, PELOTA, Uncharacterised

Background humans [1]. In the pathogenic strains, novel genetic Escherichia coli is one of the most abundant, facultative islands and small clusters of genes are present in anaerobic gram-negative bacterium of the intestinal addition to the core genomic framework and provide the microflora and colonises the mucus layer of the colon. bacteria with increased virulence [2–4]. The extracellular The core genomic structure is common among the com- intestinal pathogen, enterohemorrhagic E. coli (EHEC), mensal strains and the various pathogenic E. coli strains which cause diarrhea, hemorrhagic colitis and the that cause intestinal and extra-intestinal diseases in haemolytic uremic syndrome, is the most devastating of the pathogenic E. coli strains [5, 6]. Pathogenic bacteria have evolved various strategies to * Correspondence: [email protected] National Centre for Biological Sciences, Tata Institute of Fundamental counteract host defences. They are also exposed to Research, Bellary Road, Bangalore, Karnataka 560 065, India

© The Author(s). 2017 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 2 of 23

environments that are undergoing constant changes. Methods Hence, in order to survive, bacteria must adapt them- Dataset selves to the changing environmental conditions by al- Protein families were grouped on the basis of either tering gene expression levels and in turn adjusting structural homology (structure-centric families) or se- protein levels according to the need of the cell. Such quence homology (sequence-centric families). A dataset regulations may occur at the transcriptional and/or post- of 1285 RNA-protein and 14 DNA/RNA hybrid-protein transcriptional levels [7]. complexes were collected from the Protein Data Bank RNA-binding proteins (RBPs) are a versatile group of (PDB) (May 2015) and were split into protein and RNA proteins that perform a diverse range of functions in the chains. The RNA-interacting protein chains in this data- cell and are ‘master regulators’ of co-transcriptional and set were classified into 182 Structural Classification of post-transcriptional gene expression like RNA modifi- Proteins (SCOP) families, 135 clustered families and 127 cation, export, localization, mRNA translation, turnover orphan families (a total of 437 structure-centric fam- [8–12] and also aid in the folding of RNA into con- ilies), on the basis of structural homology with each formations that are functionally active [13]. In bacteria, other. Sequence-centric RNA-binding families were many different classes of RBPs interact with small RNAs retrieved from Pfam, using an initial keyword search of (sRNA) to form ribonucleoprotein (RNP) complexes that ‘RNA’, followed by manual curation to generate a dataset participate in post-transcriptional gene regulation pro- of 746 families. The structure-centric classification cesses [14–23]. In eukaryotes, noncoding RNAs (ncRNAs) scheme, the generation of structure-centric family are known to be important regulators of gene expression Hidden Markov Models (HMMs) and retrieval of [24–26]. Hence, bacterial RBPs that are capable of inhibit- sequence-centric family HMMs from the Pfam database ing this class of RNAs, are also capable of disrupting the (v 28) were as adapted from our previous study [43]. normal functioning of their host cells, thus acting as vi- Proteomes of 19 E. coli strains were retrieved from rulence factors. Roles of RBPs like the Hfq [27–36], Re- UniProt Proteomes (May 2016) [44] for the comparative pressor of secondary metabolites A (RsmA) [36–41] and study of pathogenic and non-pathogenic strains. The endoribonuclease YbeY [42] as virulence factors, have also names and organism IDs of the E. coli strains, their cor- been very well studied. responding UniProt proteome IDs and the total number Here, we describe the employment of mathematical of proteins in each proteome have been listed in Table 1. profiles of RBP families to study the RBP repertoire, All complete E. coli proteomes were retrieved from henceforth referred to as the ‘RBPome’,inE. coli RefSeq (May 2016) [45] to study the overall RBP strains. The proteomes of 19 E. coli strains (16 patho- landscape in E. coli.ThenamesoftheE. coli strains, genic and three non-pathogenic strains) have been stu- their corresponding assembly IDs and the total num- died to compare and contrast the RBPomes of pathogenic ber of proteins in each proteome and have been listed and non-pathogenic E. coli. More than 40 different kinds in Table 2. of proteins have been found to be present in two or more pathogenic strains, but absent from all the three non- pathogenic ones. Many of these proteins are previously Search method uncharacterised and may be novel virulence factors and The search method was described in our previous study probable candidates for further experimental validations. [43] and is represented schematically in Fig. 1. A library We have also extended our search method to probe to of 1183 RBP family HMMs (437 structure-centric fam- all available E. coli complete proteomes (till the date of ilies and 746 sequence-centric families) were used as the study) for RBPs, and thus obtain a bigger picture of start points to survey the E. coli proteomes for the pres- the RBP landscape in all known E. coli strains. The ence of putative RBPs. The genome-wide survey (GWS) search method can also be adapted in future for compar- for each E. coli proteome was performed with a se- − ing the RBPomes of other species of bacteria as well. In quence E-value cut-off of 10 3 and the hits were filtered addition, our work also discusses case studies on a few with a domain i-Evalue cut-off of 0.5. i-Evalue (inde- interesting RBPs. The first of them is an attempt to pro- pendent E-value) is the E-value that the sequence/profile vide a structural basis for the inactivity of the Ribonu- comparison would have received if this were the only clease PH (RNase PH) protein from E. coli strain K12, domain envelope found in it, excluding any others. This the second study deals with the structural modelling and is a stringent measure of how reliable this particular characterisation of RNA substrates of an ‘uncharac- domain may be. The independent E-value uses the total terised’ protein that is exclusively found in the patho- number of targets in the target database. We have now genic E. coli strains, whereas the third one involves the mentioned this definition in the revised manuscript. The analysis of pathogen-specific Cas6 proteins and compari- Pfam (v 28) domain architectures (DAs) were also resolved son with their non-pathogenic counterparts. at the same sequence E-value and domain i-Evalue cut-offs. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 3 of 23

Table 1 E. coli proteomes for comparative study. The 19 E. coli proteomes from UniProt (May 2016) used in the study for the comparison of RBPomes of pathogenic and non-pathogenic strains have been listed in this table. The pathogenic and the non- pathogenic E. coli strains have been represented in red and green fonts, respectively

apoints to the differences among the strains at the genome level, considering the strain K12 (or lab strain) as the standard

− Comparison of RNA-binding proteins across strains cut-off of 10 5. The BLASTP hits were also clustered on The RBPs identified from 19 different strains of E. coli, the basis of 100% sequence identity, 100% query coverage were compared by performing all-against-all protein and equal length cut-offs to identify identical proteins. sequence homology searches using the BLASTP module Clusters that consist of proteins from two or more of the of the NCBI BLAST 2.2.30 + suite [46] with a sequence pathogenic strains, but not from any of the non- − E-value cut-off of 10 5. The hits were clustered on the pathogenic ones, will henceforth be referred as ‘pathogen- basis of 30% sequence identity and 70% query coverage specific clusters’ and the proteins in such clusters as cut-offs to identify similar proteins i.e., proteins that had ‘pathogen-specific proteins’. Sequence homology searches a sequence identity of greater than or equal to 30%, as were performed for these proteins against the reference well as a query coverage of greater than or equal to 70%, human proteome (UP000005640) retrieved from Swiss- − were considered to homologous in terms of sequence Prot (June 2016) [44] at a sequence E-value cut-off of 10 5. and hence clustered. These parameters were standar- The hits were filtered on the basis of 30 percentage se- dised on the basis of previous work from our lab to quence identity and 70 percentage query coverage cut-offs. identify true positive sequence homologues [47]. Associations for proteins that were annotated as ‘hypo- Modelling and dynamics studies of RNase PH protein thetical’ or ‘uncharacterised,’ were obtained by sequence The structures of the active and inactive monomers of homology searches against the NCBI non-redundant (NR) the tRNA processing enzyme Ribonuclease PH (RNase protein database (February 2016) with a sequence E-value PH) from strains O26:H11 (UniProt ID: C8TLI5) and Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 4 of 23

Table 2 Complete E. coli proteomes. The 166 E. coli complete proteomes from RefSeq (May 2016) that have been used in the study have been listed in this table Organism/Name Strain Assembly Proteins E. coli O157:H7 str. Sakai Sakai substr. RIMD 0509952 GCA_000008865.1 5292 E. coli IAI39 IAI39 GCA_000026345.1 4725 E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_000005845.2 4140 E. coli O83:H1 str. NRG 857C NRG 857C GCA_000183345.1 4582 E. coli O104:H4 str. 2011C-3493 2011C-3493 GCA_000299455.1 5149 E. coli CFT073 CFT073 GCA_000007445.1 4897 E. coli BL21(DE3) BL21(DE3) GCA_000009565.2 4302 E. coli str. K-12 substr. W3110 K-12 substr. W3110 GCA_000010245.1 4410 E. coli SE11 SE11 GCA_000010385.1 4968 E. coli SE15 SE15 GCA_000010485.1 4573 E. coli O103:H2 str. 12,009 12,009 GCA_000010745.1 5423 E. coli O111:H- str. 11,128 11,128 GCA_000010765.1 5673 E. coli UTI89 UTI89 GCA_000013265.1 4963 E. coli 536 536 GCA_000013305.1 4542 E. coli APEC O1 APEC O1 GCA_000014845.1 5292 E. coli O139:H28 str. E24377A E24377A GCA_000017745.1 5021 E. coli HS HS GCA_000017765.1 4366 E. coli B str. REL606 REL606 GCA_000017985.1 4344 E. coli ATCC 8739 ATCC 8739 GCA_000019385.1 4434 E. coli str. K-12 substr. DH10B K-12 substr. DH10B GCA_000019425.1 4450 E. coli SMS-3-5 SMS-3-5 GCA_000019645.1 4908 E. coli O157:H7 str. EC4115 EC4115 GCA_000021125.1 5631 E. coli O157:H7 str. TW14359 TW14359 GCA_000022225.1 5537 E. coli BW2952 K-12 substr. BW2952 GCA_000022345.1 4347 E. coli BL21(DE3) BL21(DE3) GCA_000022665.2 4302 E. coli DH1 DH1 GCA_000023365.1 4369 E. coli ‘BL21-Gold(DE3)pLysS AG’ BL21-Gold(DE3)pLysS AG GCA_000023665.1 4322 E. coli O55:H7 str. CB9615 CB9615 GCA_000025165.1 5262 E. coli IHE3034 IHE3034 GCA_000025745.1 4911 E. coli 55,989 55,989 GCA_000026245.1 4953 E. coli IAI1 IAI1 GCA_000026265.1 4450 E. coli S88 S88 GCA_000026285.1 4696 E. coli O127:H6 str. E2348/69 E2348/69 GCA_000026545.1 4924 E. coli 042 42 GCA_000027125.1 5131 E. coli O26:H11 str. 11,368 11,368 GCA_000091005.1 5833 E. coli KO11FL KO11 GCA_000147855.3 4850 E. coli ABU 83972 ABU 83972 GCA_000148365.1 4862 E. coli UM146 UM146 GCA_000148605.1 4779 E. coli W W GCA_000184185.1 4825 E. coli ETEC H10407 ETEC H10407 GCA_000210475.1 5124 E. coli UMNK88 UMNK88 GCA_000212715.2 5542 E. coli NA114 NA114 GCA_000214765.2 4720 E. coli PCN033 PCN033 GCA_000219515.3 4881 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 5 of 23

Table 2 Complete E. coli proteomes. The 166 E. coli complete proteomes from RefSeq (May 2016) that have been used in the study have been listed in this table (Continued) E. coli UMNF18 UMNF18 GCA_000220005.2 5521 E. coli O7:K1 str. CE10 CE10 GCA_000227625.1 5152 E. coli str. ‘clone D i2’ clone D i2 GCA_000233875.1 4740 E. coli str. ‘clone D i14’ clone D i14 GCA_000233895.1 4742 E. coli O55:H7 str. RM12579 RM12579 GCA_000245515.1 5213 E. coli P12b P12b GCA_000257275.1 4549 E. coli KO11FL KO11FL GCA_000258025.1 4732 E. coli W W GCA_000258145.1 4831 E. coli Xuzhou21 Xuzhou21 GCA_000262125.1 5402 E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_000269645.2 4405 E. coli DH1 DH1 GCA_000270105.1 4351 E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_000273425.1 4404 E. coli LF82 LF82 GCA_000284495.1 4544 E. coli O25b:H4-ST131 EC958 GCA_000285655.3 5037 E. coli O104:H4 str. 2009EL-2050 2009EL-2050 GCA_000299255.1 5283 E. coli O104:H4 str. 2009EL-2071 2009EL-2071 GCA_000299475.1 5227 E. coli APEC O78 APEC O78 GCA_000332755.1 4598 E. coli str. K-12 substr. MDS42 K-12 substr. MDS42 GCA_000350185.1 3713 E. coli LY180 LY180 GCA_000468515.1 4586 E. coli PMV-1 PMV-1 GCA_000493595.1 5100 E. coli JJ1886 JJ1886 GCA_000493755.1 5151 E. coli str. K-12 substr. MC4100 K-12 substr. MC4100 GCA_000499485.1 4284 E. coli O145:H28 str. RM13514 RM13514 GCA_000520035.1 5524 E. coli O145:H28 str. RM13516 RM13516 GCA_000520055.1 5354 E. coli ST540 GCA_000597845.1 4498 E. coli ST540 GCA_000599625.1 4532 E. coli ST540 GCA_000599645.1 4550 E. coli ST2747 GCA_000599665.1 4665 E. coli ST2747 GCA_000599685.1 4585 E. coli ST2747 GCA_000599705.1 4547 E. coli O145:H28 str. RM12761 RM12761 GCA_000662395.1 5349 E. coli O145:H28 str. RM12581 RM12581 GCA_000671295.1 5520 E. coli Nissle 1917 Nissle 1917 GCA_000714595.1 4990 E. coli KLY KLY GCA_000725305.1 4478 E. coli O157:H7 str. SS17 SS17 GCA_000730345.1 5532 E. coli O157:H7 str. EDL933 EDL933 GCA_000732965.1 5530 E. coli ATCC 25922 ATCC 25922 GCA_000743255.1 4940 E. coli BW25113 K-12 substr. BW25113 GCA_000750555.1 4398 E. coli ECONIH1 GCA_000784925.1 5320 E. coli ER2796 ER2796 GCA_000800215.1 4311 E. coli K-12 ER3413 GCA_000800765.1 4309 E. coli RS218 RS218 GCA_000800845.2 4791 E. coli RM9387 GCA_000801165.1 4775 E. coli 94–3024 GCA_000801185.2 4792 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 6 of 23

Table 2 Complete E. coli proteomes. The 166 E. coli complete proteomes from RefSeq (May 2016) that have been used in the study have been listed in this table (Continued) E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_000801205.1 4387 E. coli O157:H7 str. SS52 SS52 GCA_000803705.1 5489 E. coli APEC IMT5155 APEC IMT5155 GCA_000813165.1 4840 E. coli 6409 GCA_000814145.2 4893 E. coli GCA_000819645.1 4996 E. coli O157:H16 Santai GCA_000827105.1 4776 E. coli 1303 1303 GCA_000829985.1 4849 E. coli C41(DE3) GCA_000830035.1 4302 E. coli ECC-1470 ECC-1470 GCA_000831565.1 4673 E. coli BL21 (TaKaRa) GCA_000833145.1 4262 E. coli MNCRE44 GCA_000931565.1 5137 E. coli K-12 substr. RV308 GCA_000952955.1 4342 E. coli K-12 substr. HMS174 GCA_000953515.1 4344 E. coli HUSEC2011 GCA_000967155.1 5294 E. coli VR50 VR50 GCA_000968515.1 4968 E. coli CI5 GCA_000971615.1 4874 E. coli K-12 ER3454 GCA_000974405.1 4375 E. coli K-12 ER3440 GCA_000974465.1 4367 E. coli K-12 ER3476 GCA_000974505.1 4354 E. coli K-12 ER3445 GCA_000974535.1 4360 E. coli K-12 ER3466 GCA_000974575.1 4415 E. coli K-12 ER3446 GCA_000974825.1 4357 E. coli K-12 ER3475 GCA_000974865.1 4359 E. coli K-12 ER3435 GCA_000974885.1 4443 E. coli K-12 K-12 substr. AG100 GCA_000981485.1 4394 E. coli O104:H4 str. C227–11 C227–11 GCA_000986765.1 5269 E. coli SEC470 GCA_000987875.1 4941 E. coli SQ37 GCA_000988355.1 4405 E. coli SQ88 GCA_000988385.1 4403 E. coli SQ2203 GCA_000988465.1 4402 E. coli CFSAN029787 GCA_001007915.1 5090 E. coli K-12 K-12 substr. GM4792 GCA_001020945.2 4362 E. coli K-12 K-12 substr. GM4792 GCA_001021005.2 4368 E. coli PCN061 PCN061 GCA_001029125.1 4680 E. coli C43(DE3) GCA_001039415.1 4254 E. coli NCM3722 GCA_001043215.1 4530 E. coli ACN001 ACN001 GCA_001051135.1 4671 E. coli DH1Ec095 GCA_001183645.1 4345 E. coli DH1Ec104 GCA_001183665.1 4342 E. coli DH1Ec169 GCA_001183685.1 4342 E. coli RR1 GCA_001276585.1 4337 E. coli SF-088 GCA_001280325.1 5019 E. coli SF-468 GCA_001280345.1 5218 E. coli SF-166 GCA_001280385.1 4773 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 7 of 23

Table 2 Complete E. coli proteomes. The 166 E. coli complete proteomes from RefSeq (May 2016) that have been used in the study have been listed in this table (Continued) E. coli SF-173 GCA_001280405.1 4936 E. coli O157:H7 WS4202 GCA_001307215.1 5294 E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_001308065.1 4398 E. coli K-12 substr. MG1655_TMP32XR1 GCA_001308125.1 4398 E. coli K-12 substr. MG1655_TMP32XR2 GCA_001308165.1 4399 E. coli 2012C-4227 GCA_001420935.1 5142 E. coli 2009C-3133 GCA_001420955.1 5311 E. coli YD786 GCA_001442495.1 4604 E. coli CQSW20 GCA_001455385.1 4142 E. coli uk_P46212 GCA_001469815.1 5023 E. coli ST648 GCA_001485455.1 4838 E. coli CD306 GCA_001513615.1 5021 E. coli JJ2434 GCA_001513635.1 5099 E. coli ACN002 GCA_001515725.1 4618 E. coli MRE600 GCA_001542675.2 4603 E. coli str. K-12 substr. MG1655 K-12 substr. MG1655 GCA_001544635.1 4407 E. coli JEONG-1266 GCA_001558995.1 5358 E. coli B C2566 GCA_001559615.1 4209 E. coli B C3029 GCA_001559635.1 4317 E. coli K-12 DHB4 GCA_001559655.1 4522 E. coli K-12 C3026 GCA_001559675.1 4731 E. coli str. K-12 substr. MG1655 JW5437–1 substr. MG1655 GCA_001566335.1 4405 E. coli SaT040 GCA_001566615.1 4963 E. coli G749 GCA_001566635.1 4983 E. coli ZH193 GCA_001566675.1 5040 E. coli ZH063 GCA_001577325.1 4984 E. coli JJ1887 JJ1887 GCA_001593565.1 5142 E. coli str. Sanji Sanji GCA_001610755.1 5172 E. coli 28RC1 GCA_001612475.1 5504 E. coli SRCC 1675 GCA_001612495.1 5511 E. coli Ecol_732 GCA_001617565.1 5243 E. coli Ecol_743 GCA_001618325.1 4866 E. coli Ecol_745 GCA_001618345.1 4803 E. coli Ecol_448 GCA_001618365.1 4956 E. coli B7A B7A GCA_000725265.1 5384

K12 (UniProt ID: P0CG19), respectively, were modelled HARMONY [52]. The best model for each of the active on the basis of the RNase PH protein from Pseudomonas and inactive RNase PH monomers were selected on the aeruginosa (PDB code: 1R6M: A) (239 amino acids) basis of Discrete Optimized Protein Energy (DOPE) using the molecular modelling program MODELLER v score and other validation parameters obtained from the 9.15 [48]. The active and inactive RNase PH monomers above-mentioned programs. The best models for the ac- are 238 and 228 amino acids in length, respectively and tive and inactive RNase PH monomers were subjected to are 69% and 70% identical to the template, respectively. 100 iterations of the Powell energy minimisation method Twenty models were generated for each of the active in the Tripos Force Field (in absence of any electrostat- and inactive RNase PH monomers and validated using ics) using SYBYL7.2 (Tripos Inc.). These were subjected PROCHECK [49], VERIFY3D [50], ProSA [51] and to 100 ns (ns) molecular dynamics (MD) simulations Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 8 of 23

Fig. 1 Search scheme for the genome-wide survey. A schematic representation of the search method for the GWS has been represented in this figure. Starting from 437 structure-centric and 746 sequence-centric RBP families, a library of 1183 RBP family HMMs were built. These mathematical profiles were then used to search proteomes of 19 different E. coli strains (16 pathogenic and three non-pathogenic strains). It is to be noted here that the same search scheme has been used later to extend the study to all 166 available E. coli proteomes in the RefSeq database as of May 2016 (see text for further details)

(three replicates each) in the AMBER99SB protein, nucleic acids) was modelled on the basis of the L7Ae protein from AMBER94 force field [53] using the Groningen Machine Methanocaldococcus jannaschii (PDB code: 1XBI: A) (117 for Chemical Simulations (GROMACS 4.5.5) program [54]. amino acids) and validated, as described earlier. The 64 The biological assembly (hexamer) of RNase PH from amino acids long PELOTA_1 domain of the uncharac- Pseudomonas aeruginosa (PDB code: 1R6M) served as terised protein, has 36% sequence identity with the corre- the template and was obtained using the online tool sponding 75 amino acids domain of the template. The (PISA) (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html) best model was selected as described in the case study on [55]. The structures of the active and inactive hexamers RNase PH. This model was subjected to 100 iterations of of RNase PH from strains O26:H11 and K12, respectively the Powell energy minimisation method in the Tripos were modelled and the 20 models generated for each of Force Field (in absence of any electrostatics) using the active and inactive RNase PH hexamers were validated SYBYL7.2 (Tripos Inc.). Structural alignment of the mod- using the same set of tools, as mentioned above. The best elled PELOTA_1 domain and the L7Ae K-turn binding models were selected and subjected to energy minimisa- domain from Archaeoglobus fulgidus (PDB code: 4BW0: tions, as described above. Electrostatic potential on the B) was performed using Multiple Alignment with Transla- solvent accessible surfaces of the proteins were calculated tions and Twists (Matt) [59]. The same kink-turn RNA using PDB2PQR [56] (in the AMBER force field) and from H. marismortui, found in complex with the L7Ae Adaptive Poisson-Boltzmann Solver (APBS) [57]. The K-turn binding domain from A. fulgidus, was docked onto head-to-head dimers were randomly selected from both the model, guided by the equivalents of the RNA- the active and the inactive hexamers of the protein for interacting residues (at a 5 Å cut-off distance from the performing MD simulations, to save computational time. protein) in the A. fulgidus L7Ae protein (highlighted in Various energy components of the dimer interface were yellow in the upper panel of Fig. 7c) using the molecular measured using the in-house algorithm, PPCheck [58]. docking program HADDOCK [60]. The model and the This algorithm identifies interface residues in protein- L7Ae protein from A. fulgidus, in complex with kink-turn protein interactions on the basis of simple distance RNA from H. marismortui, were subjected to 100 ns MD criteria, following which the strength of interactions at the simulations (three replicates each) in the AMBER99SB interface are quantified. 100 ns MD simulations (three protein, nucleic AMBER94 force field using the GRO- replicates each) were performed with the same set of pa- MACS 4.5.5 program. rameters as mentioned above for the monomeric proteins. Sequence analysis of pathogen-specific Cas6-like proteins Modelling and dynamics studies of an ‘uncharacterised’ The sequences of all the proteins in Cluster 308 were pathogen-specific protein aligned to the Cas6 protein sequence in E. coli strain The structure of the PELOTA_1 domain (Pfam ID: K12 (UniProt ID: Q46897), using MUSCLE [61] and PF15608) of an ‘uncharacterised’ pathogen-specific protein subjected to molecular phylogeny analysis using the from strain O103:H2 (UniProt ID: C8TX32) (371 amino Maximum Likelihood (ML) method and a bootstrap Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 9 of 23

value of 1000 in MEGA7 (CC) [62, 63]. All reviewed criteria, from the associated crRNAs (PDB codes: 4QYZ: CRISPR-associated Cas6 protein sequences were also re- L, 5H9E: L and 5H9F: L, respectively) and the protein trieved from Swiss-Prot (March 2017) [44], followed by chains (PDB codes: 4QYZ: A-J, 5H9E: A-J and 5H9F: manual curation to retain 18 Cas6 proteins. Sequences A-J, respectively), respectively. of two uncharacterised proteins (UniProt IDs: C8U9I8 and C8TG04) from Cluster 308, known to be homolo- Results gous to known CRISPR-associated Cas6 proteins (on the Genome-wide survey (GWS) of RNA-binding proteins in basis of sequence homology searches against the NR pathogenic and non-pathogenic E. coli strains database, as described earlier) were aligned to those of The GWS of RBPs was performed in 19 different E. coli the 18 reviewed Cas6 proteins using MUSCLE. The strains (16 pathogenic and three non-pathogenic strains) sequences were then subjected to molecular phylogeny and a total of 7902 proteins were identified (Additional analysis using the above-mentioned parameters. Secon- file 1: Table S1). Figure 2a shows the number of RBPs dary structure predictions for all the proteins were found in each of the strains studied here. The patho- performed using PSIPRED [64]. genic strains have a larger RBPome, as compared to the The structures of Cas6 proteins from E. coli strain K12 non-pathogenic ones - with strain O26:H11 encoding (PDB codes: 4QYZ: K, 5H9E: K and 5H9F: K) were the greatest (441). The pathogenic strains also have big- retrieved from the PDB. The RNA-binding and protein- ger proteome sizes (in terms of the number of proteins interacting residues in the Cas6 protein structures were in the proteome), as compared to their non-pathogenic calculated on the basis of 5 Å and 8 Å distance cut-off counterparts, by virtue of maintaining plasmids in them.

Fig. 2 Statistics for the genome-wide survey of 19 E. coli strains. The different statistics obtained from the GWS have been represented in this figure. In panels a and b, the pathogenic strains have been represented in red and the non-pathogenic ones in green. The non-pathogenic strains have also been highlighted with green boxes. a. The number of RBPs in each strain. The pathogenic O26:H11 strain encodes the highest number of RBPs in its proteome. b. The percentage of RBPs in the proteome of each strain. These percentages have been calculated with respect to the proteome size of the strain under consideration. The difference in this number among the pathogenic and the non-pathogenic strains are insignificant (Welch Two Sample t-test: t = 3.2384, df = 2.474, p-value = 0.06272). c. The type of Pfam domains encoded by each strain. The difference in the types of Pfam domains, as well as Pfam RBDs, encoded by the pathogenic and the non-pathogenic strains are insignificant (Welch Two Sample t-test for types of Pfam domains: t = −1.3876, df = 2.263, p-value = 0.2861; Welch Two Sample t-test for types of Pfam RBDs: t = −0.9625, df = 2.138, p-value = 0.4317). d. The abundance of Pfam RBDs. 185 types of Pfam RBDs were found to be encoded in the RBPs, of which DEAD domains have the highest representation (approximately 4% of all Pfam RBDs) Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 10 of 23

Hence, to normalise for proteome size, the number of cassette (ABC) transporters [66]. The ATP-binding sub- RBPs in each of these strains were expressed as a function unit of ABC transporters share common features with of their respective number of proteins in the proteome other nucleotide-binding proteins [67] like, the E. coli (Fig. 2b). We observed that the difference in the percent- RecA [68] and the F1-ATPase from bovine heart [69]. age of RBPs in the proteome among the pathogenic and GCN20, YEF3 and RLI1 are examples of soluble ABC the non-pathogenic strains are insignificant (Welch Two proteins that interact with ribosomes and regulate trans- Sample t-test: t = 3.2384, df = 2.474, p-value = 0.06272). lation and ribosome biogenesis [70–72]. To compare the differential abundance of domains, if The other large MMCs were those of small toxic poly- any, among the pathogens and the non-pathogens, the peptides that are components of the bacterial toxin- Pfam DAs of all the RBPs were resolved (to strengthen antitoxin (TA) systems [73–77], RNA helicases that are the results in this section, this study has been extended involved in various aspects of RNA metabolism [78, 79] to all known E. coli proteomes and will be discussed in a and Pseudouridine synthases that are enzymes respon- later section). The number of different types of Pfam sible for pseudouridylation, which is the most abundant domains and that of Pfam RNA-binding domains (RBDs) post-transcriptional modification in RNAs [80]. Cold found in each strain have been represented in Fig. 2c. shock proteins bind mRNAs and regulate translation, We observed that the difference in the types of Pfam rate of mRNA degradation etc. [81, 82]. These proteins domains, as well as Pfam RBDs, encoded by the patho- are induced during the response of the bacterial cell genic and the non-pathogenic strains are insignificant towards temperature rise. (Welch Two Sample t-test for types of Pfam domains: The majority of the SMCs (38 out of 48 SMCs) are t = − 1.3876, df = 2.263, p-value = 0.2861; Welch Two RBPs from pathogenic strains and lack homologues in Sample t-test for types of Pfam RBDs: t = − 0.9625, df any of the other strains considered here. These include = 2.138, p-value = 0.4317). The number of different proteins like putative helicases, serine proteases, and Pfam RBDs, found across all the 19 E. coli strains various endonucleases. Likewise, members of the small studied here, has been shown in Fig. 2d and also toxic Ibs protein family (IbsA, IbsB, IbsC, IbsD and been listed in Table 3. IbsE that form Clusters 362, 363, 364, 365 and 366 We found that E. coli encodes 185 different types of respectively) from strain K12 are noteworthy examples Pfam RBDs in their proteomes and the DEAD domain of SMCs that are in non-pathogenic strains only. These was found to be the most abundant, constituting Ibs proteins cause the cessation of growth when over- approximately 4% of the total number of Pfam RBD expressed [83]. domains in E. coli. The DEAD box family of proteins are RNA helicases that are required for RNA metabolism and thus are important players in gene expression [65]. Pathogen-specific proteins These proteins use ATP to unwind short RNA duplexes In this study, the 226 pathogen-specific proteins that in an unusual fashion and also help in the remodelling formed 43 pathogen-specific clusters are of special inter- of RNA-protein complexes. est. Sixty-three of these proteins were previously unchar- acterised and associations for all of these proteins were Comparison of RNA-binding proteins across strains obtained on the basis of sequence homology searches reveals novel pathogen-specific factors against the NCBI-NR database. The function annotation The proteins were clustered on the basis of sequence of each of these clusters were transferred on the basis of homology searches in order to compare and contrast homology. The biological functions and the number of the RBPs across the E. coli strains studied here. The RBPs constituting these pathogen-specific clusters have 7902 proteins identified from all the strains were been listed in Table 4. grouped into 384 clusters, on the basis of sequence hom- If these pathogen-specific proteins are exclusive to the ology with other members of the cluster (Additional file 2: pathogenic strains, then they may be exploited for drug Table S2). Greater than 99% of the proteins could cluster design purposes. To test this hypothesis, we surveyed with one or more RBPs and formed 336 multi-member the human (host) proteome for the presence of sequence clusters (MMCs), whereas the rest of the proteins failed to homologues of these proteins. It was found that, barring cluster with other RBPs and formed 48 single-member the protein kinases that were members of Cluster 98 clusters (SMCs). The distribution of members among all (marked in asterisk in Table 4), none of the pathogen- the 384 clusters has been depicted in Fig. 3. specific proteins were homologous to any human protein The largest of the MMCs, consists of 1459 RBPs which within the thresholds employed in the search strategy are ATP-binding subunit of transporters. The E. coli (please see Methods section for details). Few of the genome sequence had revealed that the largest family of pathogen-specific protein clusters are described in the paralogous proteins were composed of ATP-binding following section. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 11 of 23

Table 3 Pfam RNA-binding domains. The Pfam RBDs and their corresponding occurrences in the GWS of 19 E. coli strains have been listed in this table. The Pfam domains listed are on the basis of Pfam database (v.28) Pfam domain Number of occurrences Pfam domain Number of occurrences DEAD 250 MnmE_helical 19 S1 247 RNA_pol_Rpb2_7 19 GTP_EFTU 168 tRNA-synt_2e 19 GTP_EFTU_D2 168 Ribosomal_L11_N 19 HOK_GEF 160 KilA-N 19 S4 152 Ribosomal_S4 19 PseudoU_synth_2 150 RTC 19 CSD 133 Ribosomal_L2 19 SpoU_methylase 114 DnaB_bind 19 tRNA_anti-codon 113 IPPT 19 RNase_T 99 IF3_C 19 tRNA-synt_1 95 UPF0020 19 tRNA-synt_2 94 RNA_pol_A_bac 19 Ribonuc_L-PSP 91 RNA_pol_Rpb1_3 19 PRD 90 Se-cys_synth_N 19 Anticodon_1 76 RimM 19 Formyl_trans_N 75 Val_tRNA-synt_C 19 Aconitase 75 TruB_C_2 19 RF-1 70 RNA_pol_Rpb1_4 19 tRNA-synt_1c 58 dsrm 19 RNase_PH 57 RNA_pol_Rpb2_1 19 RNase_PH_C 57 Rho_N 19 Dus 57 CheR 19 HGTP_anticodon 57 SHS2_FTSA 19 tRNA-synt_2b 57 Helicase_RecD 19 tRNA_bind 56 RNA_pol_Rpb6 19 Sua5_yciO_yrdC 56 RNA_pol_Rpb2_6 19 tRNA_U5-meth_tr 56 SpoU_methylas_C 19 zf-FPG_IleRS 55 Trm112p 19 Ldr_toxin 52 RNA_pol_Rpb1_1 19 RtcB 50 CsrA 19 CorA 46 Ribosomal_S20p 19 MazE_antitoxin 44 TruB-C_2 19 ProQ 41 DALR_2 19 DbpA 39 IF-2 19 HA2 39 tRNA_Me_trans 19 PolyA_pol_RNAbd 38 KH_1 19 TruD 38 ABC1 19 IF2_N 38 tRNA-synt_1_2 19 PseudoU_synth_1 38 PUA 19 Methyltr_RsmF_N 38 CAT_RBD 19 SpoU_sub_bind 38 PNPase 19 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 12 of 23

Table 3 Pfam RNA-binding domains. The Pfam RBDs and their corresponding occurrences in the GWS of 19 E. coli strains have been listed in this table. The Pfam domains listed are on the basis of Pfam database (v.28) (Continued) SgrR_N 38 tRNA_m1G_MT 19 tRNA_edit 38 SelB-wing_2 19 PolyA_pol 38 RNase_H 19 FtsJ 38 PRC 19 HRDC 38 Ribosomal_L18p 19 tRNA-synt_1b 38 GIDA_assoc 19 THUMP 38 RrnaAD 19 DALR_1 38 YjeF_N 19 NusB 38 LigT_PEase 19 RNase_E_G 38 Ub-RnfH 19 ASCH 37 Nol1_Nop2_Fmu_2 19 DNA_pol_A_exo1 37 RRF 19 tRNA_SAD 36 B5 18 DHHA1 36 FDX-ACB 18 GTP_EFTU_D3 35 GidB 18 MqsA_antitoxin 27 Sigma70_r1_1 18 TGT 21 IF3_N 18 RNA_pol_Rpb1_2 20 B3_4 18 Queuosine_synth 20 tRNA-synt_2c 17 RVT_1 20 Colicin-DNase 16 tRNA_synt_2f 19 PTS_2-RNA 16 SelB-wing_3 19 CRISPR_Cse1 14 Methyltrans_RNA 19 CRISPR_Cse2 14 Ribosomal_L11 19 FinO_N 13 CRS1_YhbY 19 PIN 13 Tyr_Deacylase 19 GIIM 11 Ribosomal_L4 19 IlvGEDA_leader 11 MutS_II 19 SymE_toxin 11 RNA_pol_Rpb1_5 19 PRTase_1 10 TilS 19 PELOTA_1 10 Endonuclease_1 19 DNA_primase_S 10 Ribosomal_L25p 19 IlvB_leader 9 Sigma54_CBD 19 YafO_toxin 8 RapA_C 19 MT-A70 8 TilS_C 19 RPAP2_Rtr1 6 OB_NTP_bind 19 TisB_toxin 5 GlutR_dimer 19 MqsR_toxin 5 RNA_pol_Rpb2_3 19 Ibs_toxin 5 RtcR 19 RNA_ligase 4 TruB_N 19 N36 3 RsmJ 19 Cloacin 3 tRNA_bind_2 19 Colicin_D 2 tRNA-synt_1c_C 19 Viral_helicase1 2 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 13 of 23

Table 3 Pfam RNA-binding domains. The Pfam RBDs and their corresponding occurrences in the GWS of 19 E. coli strains have been listed in this table. The Pfam domains listed are on the basis of Pfam database (v.28) (Continued) RNA_pol_Rpb2_45 19 IF-2B 1 HisG 19 Colicin_immun 1 Rho_RNA_bind 19 RPOL_N 1 PNPase_C 19 RNA_pol 1 RNase_HII 19 RnlA_toxin 1 RNA_pol_A_CTD 19 NYN 1 RNA_pol_L 19 DUF3850 1 Rsd_AlgQ 19

The DEAD/DEAH box helicases that use ATP to fertility inhibition complex which regulates the expression unwind short duplex RNA [65], formed three different of the genes in the transfer operon [86–89]. tRNA (fMet)- clusters. In two of the clusters, the DEAD domains specific endonucleases are the toxic components of a TA (Pfam ID: PF00270) were associated with C-terminal system. This site-specific tRNA-(fMet) endonuclease acts Helicase_C (Pfam ID: PF00271) and DUF1998 (Pfam ID: as a virulence factor by cleaving both charged and PF09369) domains. On the other hand, in a bigger uncharged tRNA-(fMet) and inhibiting translation. The cluster, the DEAD/DEAH box helicases were composed Activating Signal Cointergrator-1 homology (ASCH) of DNA_primase_S (Pfam ID: PF01896), ResIII (Pfam domain is also a putative RBD due to the presence of an ID: PF04851) and Helicase_C domains. Four of the RNA-binding cleft associated with a conserved sequence pathogen-specific clusters were those of Clustered Regu- motif characteristic of the ASC-1 superfamily [90]. larly Interspaced Short Palindromic Repeat (CRISPR) sequence-associated proteins, consisting of RBPs from 10 Identification of the distinct RNA-binding protein repertoire pathogenic strains each. Recent literature reports also sup- in E. coli port the role of CRISPR-associated proteins as virulence We identified identical RBPs across E. coli strains, on factors in pathogenic bacteria [84]. The KilA-N domains the basis of sequence homology searches and other are found in a wide range of proteins and may share a filtering criteria (as mentioned in the Methods section). common fold with the nucleic acid-binding modules of Out of the 7902 RBPs identified in our GWS, 6236 had certain nucleases and the N-terminal domain of the tRNA one or more identical partners from one or more strains endonuclease [85]. Fertility inhibition (FinO) protein and and formed 1227 clusters, whereas 1666 proteins had no the anti-sense FinP RNA are members of the FinOP identical counterparts. Hence, our study identified 2893 RBPs from 19 E. coli strains that were distinct from each other. Identification of such a distinct pool of RBPs will help to provide an insight to the possible range of func- tions performed by this class of proteins in E. coli, and hence compare and contrast with the possible functions performed by RBPs in other organisms.

GWS of RNA-binding proteins in all known E. coli strains We extended the above-mentioned study, by performing GWS of RBPs in 166 complete E. coli proteomes avai- lable in the RefSeq database (May 2016) and a total of 8464 proteins were identified (Additional file 3). It should be noted that, unlike the nomenclature system of Fig. 3 Clusters of RNA-binding proteins. The percentage of RBPs in UniProt, where the same protein occurring in different the different clusters has been represented in this figure. The RBPs strains are denoted with different UniProt accession IDs, obtained from each of the 19 E. coli strains (16 pathogenic and three non-pathogenic strains) have been clustered on the basis of homology RefSeq assigns same or at times different accession IDs searches (see text for further details). Five of the biggest clusters to the same protein occurring in different strains. Thus, and their identities are as follows: Cluster 5 (ATP-binding subunit on the basis of unique accession IDs, 8464 RBPs were of transporters), Cluster 41 (Small toxic polypeptides), Cluster 15 identified. The 8464 RBPs were grouped into 401 clus- (RNA helicases), Cluster 43 (Cold shock proteins) and Cluster 16 ters on the basis of sequence homology with other mem- (Pseudouridine synthases) bers of the cluster. We found that greater than 99% of Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 14 of 23

Table 4 Pathogen-specific RNA-binding protein clusters. The Table 4 Pathogen-specific RNA-binding protein clusters. The size of RBP clusters with members from only the pathogenic E. size of RBP clusters with members from only the pathogenic E. coli strains in our GWS of 19 E. coli strains have been listed in coli strains in our GWS of 19 E. coli strains have been listed in this table this table (Continued) Cluster number Number of Cluster name Cluster 238 2 Hypothetical proteins members Cluster 302 2 Peptidyl-arginine deiminases Cluster 283 13 KilA-N domain phage proteins Cluster 303 2 Exonucleases Cluster 299 13 tRNA(fMet)-specific VapC endonucleases Cluster 315 2 KilA-N domain phage proteins Cluster 176 10 DEAD/DEAH box helicases Cluster 320 2 Pyocin, putative colicin activity proteins Cluster 307 10 CRISPR type I-E/−associated proteins CasA/Cse1 Cluster 325 2 Hyothetical proteins Cluster 308 10 CRISPR-associated proteins Cas6/Cse3/ Cluster 327 2 KilA-N domain proteins CasE, subtype I-E Cluster 332 2 Hypothetical phage associated proteins Cluster 309 10 CRISPR type I-E/−associated proteins Cluster 334 2 Hypothetical proteins CasB/Cse2 Cluster 335 2 Chromosome partitioning ParA proteins Cluster 310 10 CRISPR-associated proteins Cas5/CasD, subtype I-E Cluster 336 2 Serine/Threonine kinases Cluster 60 10 Putative ATP/GTP-binding proteins aAll the proteins in this cluster have sequence homologues in humans Cluster 318 9 ASCH domain-containing proteins Cluster 122 8 Adenine DNA methyltransferases, the proteins could cluster with one or more RBPs and phage-associated formed 339 MMCs, whereas the rest of the proteins Cluster 305 8 Post-segregational killing toxins failed to cluster with other RBPs and formed 62 SMCs. Cluster 116 7 RNA-binding proteins The above-mentioned GWS statistics for RBP num- bers have been plotted in Fig. 4a. The number of differ- Cluster 298 6 VagC, VagC-homologs ent Pfam RBDs found across all complete E. coli Cluster 317 6 Type I restriction-modification enzymes, M subunits proteomes has been shown in Fig. 4b. Similar to the afore-mentioned results, seen from the dataset of 19 E. Cluster 319 6 Type I restriction-modification enzymes, R subunits coli proteomes, it was found that E. coli encodes 188 different types of Pfam RBDs in their proteomes and the Cluster 125 5 Helicases DEAD domain was still observed to be the most abun- Cluster 246 5 DNA-binding proteins dant, constituting approximately 6% of the total number Cluster 287 5 ATP-dependent helicases of Pfam RBD domains in E. coli. The length distribution Cluster 288 5 DEAD/DEAH box helicases of RBPs from E. coli have been plotted in Fig. 4c and Cluster 290 5 DEAD/DEAH box helicases RBPs of the length 201–300 amino acids were found to Cluster 63 5 UvrD/REP helicase-like proteins be the most prevalent. Cluster 98a 5 Protein kinases ′ ′ Cluster 161 4 2 -5 RNA ligases Identification of the complete distinct RBPome in 166 Cluster 300 4 proQ/FINO family proteins proteomes of E. coli Cluster 313 4 ATP-dependent helicases, res subunit These 8464 RBPs (please see previous section) formed of Type III restriction enzyme, SNF2 1285 clusters of two or more identical proteins, accoun- family helicases ting for 3532 RBPs, whereas the remaining 4932 RBPs Cluster 329 4 Serine/Threonine kinases were distinct from the others. Hence, 6217 RBPs, dis- Cluster 301 3 YajA proteins tinct from each other, were identified from all known E. Cluster 323 3 Putative pyridoxal phosphate-dependent coli strains, which is much greater than the number enzymes, Putative transferases (2893) found from 19 E. coli proteomes. Cluster 324 3 Sigma-54 dependent transcription It should be noted that the pathogenicity annota- regulators, Putative transcriptional regulators of NtrC family tions are not very clear for few of the 166 E. coli strains for which complete proteome information are Cluster 326 3 Leucin-rich repeat proteins available. Hence, we have performed the analysis for Cluster 330 3 Ankyrin repeat-containing domain the pathogen-specific proteins using the smaller data- proteins set of 19 proteomes, whereas all the 166 complete Cluster 172 2 Zn-dependent hydrolases, including glyoxylases, Beta-lactamases proteomes have been considered for the analysis for the complete E. coli RBPome. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 15 of 23

Fig. 4 Statistics for the genome-wide survey of 166 E. coli strains. The different statistics obtained from the GWS have been represented in this figure. a The number of RBPs as determined by different methods (see text for further details). b The abundance of Pfam RBDs. 188 types of Pfam RBDs were found to be encoded in the RBPs, of which DEAD domains have the highest representation (approximately 6% of all Pfam RBDs). c The length distribution of RBPs

Case studies from all other known RNase PH proteins from E. coli Three case studies on interesting RBPs were performed and has a truncated C-terminus. In 1993, DNA sequen- to answer some outstanding questions and have been cing studies had revealed that a GC base pair (bp) was described in the following sections. The first of the three missing in this strain from a block of five GC bps found examples, deals with a RNase PH protein that does not 43–47 upstream of the rph stop codon [94]. This one cluster with those from any of the other 165 E. coli pro- base pair deletion leads to a translation frame shift over teomes considered in this study. This protein, which the last 15 codons, resulting in a premature stop codon forms a SMC, is interesting in the biological context due (five codons after the deletion). This premature stop to its difference with the other RNase PH proteins, both codon, in turn, leads to the observed reduction in size of at the level of sequence as well as biological activity. The the RNase PH protein by 10 residues. It was also shown second case study deals with a protein that is a part of a by Jensen [94] that this protein lacks RNase PH activity. pathogen-specific cluster, in which none of the proteins Figure 5a shows a schematic representation of the DAs of are well-annotated. This protein was found to encode a the active (up) and inactive (down) RNase PH proteins, bacterial homologue of a well-known archaeo-eukaryotic with the five residues that have undergone mutations and RBD, whose RNA-binding properties are not as well the ten residues that are missing from the inactive RNase studied as its homologues. The final study involves a PH protein depicted in orange and yellow, respectively. sequence-based approach to analyse the pathogen-specific These are the residues of interest in our study. The same CRISPR-associated Cas6 proteins, and compare the same colour coding has been used both in Fig. 5a and b. with similar proteins from the non-pathogenic strains. To provide a structural basis for this possible loss of activity of the RNase PH protein from strain K12, we Case study 1: RNase PH from strain K12 is inactive due to modelled the structures of the RNase PH protein mono- a possible loss of stability of the protein mer as well as the hexamer from strains O26:H11 and RNase PH is a phosphorolytic exoribonuclease involved K12 (Fig. 5b and c). It is known in the literature that the in the maturation of the 3′-end of transfer RNAs hexamer (trimer of dimers) is the biological unit of the (tRNAs) containing the CCA motif [91–93]. The RNase RNase PH protein and that the hexameric assembly is PH protein from strain K12 was found to be distinct mandatory for the activity of the protein [95, 96]. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 16 of 23

Fig. 5 Modelling of the RNase PH proteins from two different E. coli strains. The structural modelling of the RNase PH protein has been represented in this figure. a Schematic diagram of the active (above) and the inactive (below) RNase PH proteins. The RNase PH and the RNase_PH_C domains, as defined by Pfam (v.28), have been represented in magenta and pink, respectively. The five residues that have undergone mutations due to a point deletion and the ten residues that are missing from the inactive RNase PH protein from strain K12 have been depicted in orange and yellow, respectively. These two sets of residues are the ones of interest in this study. b Model of the RNase PH monomer from strain O26:H11. The residues with the same colour codes as mentioned in panel (a), have been represented on the structure of the model. The residues that are within an 8 Å cut-off distance from the residues of interest have been highlighted in cyan (left). c Structure of the RNase PH hexamer from strain O26:H11 (left) and the probable structure of the inactive RNase PH hexamer from strain K12 (right). The dimers marked in black boxes are the ones that were randomly selected for MD simulations. d Electrostatic potential on the solvent accessible surface of the RNase PH hexamer from strain O26:H11 (left) and that of the inactive RNase PH hexamer from strain K12 (right)

The stability of both the monomer and the hexamer explain the inactive nature of the protein from strain were found to be affected in strain K12, as compared to K12. Figure 5d shows differences in the electrostatic that in strain O26:H11. The energy values have been potential on the solvent accessible surfaces of the active plotted in Fig. 6a. In both monomer and hexamer, there (left) and inactive (right) RNase PH proteins. is a reduction in stability, suggesting that the absence of To test this hypothesis for the possible loss of function C-terminal residues affects the stability of the protein, of the RNase PH protein due to loss of stability of the perhaps more than a cumulative contribution to the monomer and/or the hexamer, we performed MD simu- stability of the protein. It should be noted that since the lations to understand distortions, if any, of the monomer monomeric form of the inactive protein is less stable and a randomly selected head-to-head dimer (from the than that of its active counterpart, the hexameric assem- hexameric assembly) of both the active and the inactive bly of the inactive RNase PH protein is only a putative proteins. The dimers have been marked in black boxes one. Hence, the putative and/or unstable hexameric in Fig. 5c. Various energy components of the dimer assembly of the RNase PH protein, leads to the loss of interface, as calculated by PPCheck, have been plotted in activity of the protein. Fig. 6b. The results show that the inactive RNase PH Figure 5b shows that the residues marked in cyan (left) dimer interface is less stabilised as compared to that of are at an interacting distance of 8 Å from the residues of the active protein. The trajectories of the MD runs have interest (left). These residues marked in cyan are a sub- been shown in additional movie files (Additional file 4, set of the RNase PH domain, which is marked in ma- Additional file 5, Additional file 6 and Additional file 7, genta (right). Hence, the loss of possible interactions for the active monomer, inactive monomer, active dimer (between the residues marked in cyan and the residues and inactive dimer, respectively). Analyses of Additional of interest) and subsequently stability of the three- file 4, and Additional file 5 shows a slight distortion in dimensional structure of the RNase PH domain might the short helix (pink) in the absence of residues of Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 17 of 23

Fig. 6 Energy values for the active and inactive RNase PH monomers, dimers and hexamers. The energy values (in kJ/mol) for the active (blue) and the inactive (red) RNase PH proteins, as calculated by SYBYL (in panel a) and PPCheck (in panel b) have been plotted in this figure. a The energy values for the active and the inactive RNase PH monomers and hexamers. The results show that both the monomeric, as well as the hexameric forms of the inactive RNase PH protein, is unstable as compared to the those of the active RNase PH protein. b The interface energy values for the active and the inactive RNase PH dimers (as marked in black boxes in Fig. 5c). The results show that the dimer interface of the inactive RNase PH protein is less stabilised as compared to that of the active RNase PH protein

interest (orange and yellow), which might lead to overall as ‘putative,’ ‘uncharacterised,’ ‘hypothetical’ or ‘predicted’. loss of stability of the monomer. Further analyses (Add- To understand the RNA-binding properties of these ortho- itional file 6 and Additional file 7) show the floppy na- logous pathogen-specific proteins, we resolved the Pfam ture of the terminal part of the helices that are DA of this protein. In particular, such an association to interacting in the dimer. This is probably due to the loss Pfam domains provide function annotation to a hitherto of the residues of interest, which have been seen to be uncharacterised protein, from strain O103:H2, to RBD structured and less floppy in the active RNase PH PELOTA_1. Hence, the structure of the RNA-binding dimer (Additional file 6). PELOTA_1 domain of this protein was modelled on the For each of the systems, the H-bond traces for three basis of the L7Ae protein from M. jannaschii (Fig. 7a). replicates (represented in different colours) have been Domains that are involved in core processes, such as depicted. From these figures, we can observe that the RNA maturation, e.g. the tRNA endonucleases, and replicates are showing similar H-bonding patterns. Ana- translation and with an archaeo-eukaryotic phyletic pat- lyses of the number of hydrogen bonds (H-bonds) tern includes the PIWI, PELOTA and SUI1 domains formed in the system over each picosecond of the MD [97]. In 2014, Anantharaman and co-workers had shown simulations of the active monomer, inactive monomer, associations of the conserved C-terminus of a phosphor- active dimer and inactive dimer have been represented ibosyltransferase (PRTase) in the Tellurium resistance in Fig. 8a, b, c and d, respectively. Comparison of panels (Ter) operon to a PELOTA or Ribosomal_L7Ae domain a and b of this figure shows a greater number of (Pfam ID: PF01248) [98]. These domains are homo- H-bonds being formed in the active monomer, as com- logues of the eukaryotic release factor 1 (eRF1), which is pared to that of the inactive monomer, over the entire involved in translation termination. Unlike the well- time period of the simulation. Similarly, comparison of studied PELOTA domain, the species distribution of the panels c and d of this figure shows a greater number of PELOTA_1 domain is solely bacterial and not much is H-bonds being formed in the active dimer as compared known in literature regarding the specific function of to that of the inactive dimer, over the entire time period this domain. of the simulation. These losses of H-bonding interac- Structure of this modelled PELOTA_1 domain from tions might lead to overall loss of stability of the dimer the uncharacterised protein was aligned with that of the and subsequently that of the hexamer. L7Ae kink-turn (K-turn) binding domain from an archaeon (A. fulgidus) (Fig. 7b). The model also retained Case study 2: Uncharacterised pathogen-specific protein the same basic structural unit as the eRF1 protein (data and its homologues show subtly different RNA-binding not shown). The L7Ae is a member of a family of pro- properties teins that binds K-turns in many functional RNA species In our study, we observed that Cluster 60 was composed of [99]. The K-turn RNA was docked onto the model, 10 proteins, each from a different pathogenic strain studied guided by the equivalents of the known RNA-interacting here. All the proteins in this cluster were either annotated residues from the archaeal L7Ae K-turning binding Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 18 of 23

Fig. 7 Uncharacterised pathogen-specific RNA-binding protein. The characterisation of the uncharacterised pathogen-specific RBP has been represented in this figure. a Schematic representation of the domain architecture of the protein. The RNA-binding PELOTA_1 domain and its model has been shown here. b Structural superposition of the L7Ae K-turn binding domain (PDB code: 4BW0: B) (in red) and the model of the uncharacterised protein PELOTA_1 domain (in blue). c. Comparison of the kink-turn RNA-bound forms of the L7Ae K-turn binding domain (PDB code: 4BW0: B) (up) and that of the model of the uncharacterised protein PELOTA_1 domain (down). The RNA-binding residues have been highlighted in yellow

domain. Both the complexes have been shown in Fig. 7c Case study 3: Pathogen specific Cas6-like proteins might with the RNA-interacting residues highlighted in yellow. be functional variants of the well-characterised non- MD simulations of both these complexes were performed pathogenic protein and the trajectories have been shown in additional movie In many bacteria, as well archaea, CRISPR associated Cas files Additional file 8 (PELOTA_1 domain model-k-turn proteins and short CRISPR-derived RNA (crRNA) assem- RNA complex) and Additional file 9 (L7Ae K-turn binding ble into large RNP complexes and provide surveillance domain-k-turn RNA complex). towards invasion of genetic parasites [100–102]. The role For each of the systems, the H-bond traces for three of CRISPR-associated proteins as virulence factors in replicates (represented in different colours) have been pathogenic bacteria has also been reported in recent depicted. From these figures, one can observe that the literature [84]. We found that Cluster 308 consists of 10 replicates are showing similar H-bonding patterns. Ana- pathogen-specific proteins, of which half of them were lyses of the number of H-bonds formed between the already annotated as Cas6 proteins, whereas the other half protein and the RNA over each picosecond of the MD constituted of ‘uncharacterised’ or ‘hypothetical’ proteins. simulations of the PELOTA_1 domain-RNA complex As mentioned in the Methods section, the latter proteins and the L7Ae K-turn binding domain-RNA complex were annotated on the basis of sequence homology to have been represented in Fig. 8e and f, respectively. known proteins in the NR database, as Cas6 proteins. Comparison of panels e and f of this figure shows a Molecular phylogeny analysis of all the proteins from greater number of H-bonds being formed in the L7Ae Cluster 308 and Cas6 from E. coli strain K12 has been K-turn binding domain-RNA complex as compared to depicted in Additional file 10a: Figure S1, which rein- that of the PELOTA_1 domain-RNA complex over the states the fact that the pathogen-specific proteins are entire time period of the simulation. These results show more similar to each other, in terms of sequence, than that the two proteins have differential affinity towards they are to the Cas6 protein from the non-pathogenic the same RNA molecule. This hints at the fact that these strain K12. Furthermore, a similar analysis of two previ- proteins might perform subtly different functions by the ously uncharacterised proteins (UniProt IDs: C8U9I8 virtue of having differential RNA-binding properties. and C8TG04) (red) from this pathogen-specific Cas6 Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 19 of 23

Fig. 8 Hydrogen bonding patterns in molecular dynamics simulations. The number of H-bonds formed over each picosecond of the MD simulations (described in this Chapter) have been shown in this figure. Each of the six panels (systems) shows the H-bond traces from three replicates (represented in different colours). a Active RNase PH monomer. b Inactive RNase PH monomer. c Active RNase PH dimer. d Inactive RNase PH dimer. e PELOTA_1 domain from the ‘uncharacterised’ protein in complex with kink-turn RNA. f L7Ae K-turn binding domain from A. fulgidus in complex with kink-turn RNA from H. marismortui proteins cluster (Cluster 308), with other known Cas6 specific residues’. A similar colouring scheme has been proteins has been shown Additional file 10b: Figure S1. followed in Fig. 9b, to analyse the conservation of From the phylogenetic tree, one can infer that the protein-interacting residues in these proteins. From pathogen-specific Cas6 proteins are more similar in these analyses, we can speculate that due to the presence terms of sequence to the Cas6 from E. coli strain K12 of a large proportion of ‘class-specific residues’, the (blue) than that from other organisms. RNA-binding properties, as well as protein-protein inter- Multiple sequence alignment (MSA) of all the proteins actions, might be substantially different among the Cas6 from Cluster 308 and Cas6 from strain K12 has been proteins from non-pathogenic and pathogenic E. coli shown in Fig. 9. The RNA-binding residues in E. coli strains, which might lead to functional divergence. strain K12 Cas6 protein (union set of RNA-binding Secondary structures of each of these proteins, mapped residues inferred from each of the three known PDB on their sequence (α-helices highlighted in cyan and structures (see Methods section)) have been highlighted β-strands in green) in Fig. 9c, also hint at slight struc- in yellow on its sequence (CAS6_ECOLI) on the MSA. tural variation among these proteins. The corresponding residues in the other proteins on the MSA, which are same as that in CAS6_ECOLI, have also been highlighted in yellow, whereas those which differ Discussion have been highlighted in red. From Fig. 9a, we can We have employed a sequence search-based method to conclude that the majority of the RNA-binding residues compare and contrast the proteomes of 16 pathogenic in CAS6_ECOLI are not conserved in the pathogen- and three non-pathogenic E. coli strains as well as to specific Cas6 proteins, and can be defined as ‘class- obtain a global picture of the RBP landscape in E. coli. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 20 of 23

Fig. 9 Sequence analysis of pathogen-specific Cas6-like proteins. Comparison of sequence features of Cas6 proteins from pathogenic (Cluster 308) and non-pathogenic K12 strains. a Comparison of RNA-binding residues. The RNA-binding residues in E. coli strain K12 Cas6 protein have been highlighted in yellow on its sequence (CAS6_ECOLI) on the MSA. The corresponding residues in the other proteins on the MSA, which are same as that in CAS6_ECOLI, have also been highlighted in yellow, whereas those which differ have been highlighted in red. b Comparison of protein-interacting residues. The protein-interacting residues in E. coli strain K12 Cas6 protein have been highlighted in yellow on its sequence (CAS6_ECOLI). A similar colour scheme has also been followed here. c Secondary structure prediction. The α-helices have been highlighted in cyan and the β-strands in green

The results obtained from this study showed that the second study, a previously uncharacterised pathogen- pathogenic strains encode a greater number of RBPs in specific protein was studied and was found to possess their proteomes, as compared to the non-pathogenic subtly different RNA-binding affinities towards the same ones. The DEAD domain, involved in RNA metabolism, RNA stretch as compared to its well characterised was found to be the most abundant of all identified homologues in archaea and eukaryotes. This might hint at RBDs. The complete and distinct RBPome of E. coli was different functions of these proteins. In the third case also identified by studying all known E. coli strains till study, pathogen-specific CRISPR-associated Cas6 proteins date. In this study, we identified RBPs that were exclusive were analysed and found to have diverged functionally to pathogenic strains, and most of them can be exploited from the known prototypical Cas6 proteins. as drug targets by virtue of being non-homologous to their human host proteins. Many of these pathogen- Conclusions specific proteins were uncharacterised and their identities The approach used in our study to cross-compare pro- could be resolved on the basis of sequence homology teomes of pathogenic and non-pathogenic strains may searches with known proteins. also be extended to other bacterial or even eukaryotic Further, in this study, we performed three case studies proteomes to understand interesting differences in their on interesting RBPs. In the first of the three studies, a RBPomes. The pathogen-specific RBPs reported in this tRNA processing RNase PH enzyme from strain K12 study, may also be taken up further for clinical trials was investigated that is different from that in all other E. and/or experimental validations. coli strains in having a truncated C-terminus and being TheeffectoftheabsenceofafunctionalRNasePHinE. functionally inactive. Structural modelling and molecular coli strain K12 is not clear. The role of the PELOTA_1 dynamics studies showed that the loss of stability of the domain-containing protein may also be reinforced per- monomeric and/or the hexameric (biological unit) forms forming knockdown and rescue experiments. These might of this protein from E. coli strain K12, might be the pos- help to understand the functional overlap of this protein sible reason for the lack of its functional activity. In the with its archaeal or eukaryotic homologues. Introduction Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 21 of 23

of this pathogen-specific protein in non-pathogens might Abbreviations also provide probable answers towards its virulence pro- ABC: ATP-binding cassette transporters; APBS: Adaptive Poisson-Boltzmann Solver; ASCH: Activating Signal Cointergrator-1 homology; bp: Base pair; perties. The less conserved RNA-binding and protein- Cas: CRISPR-associated system; CRISPR: Clustered Regularly Interspaced Short interacting residues in the pathogen-specific Cas6 proteins, Palindromic Repeat; crRNA: CRISPR RNA; DA: Domain architecture; might point to functional divergence of these proteins DOPE: Discrete Optimized Protein Energy; EHEC: Enterohemorrhagic E. coli; Fin: Fertility inhibition; GROMACS: Groningen Machine for Chemical from the known ones, but warrants further investigation. Simulations; GWS: Genome-wide survey; HMM: Hidden Markov Model; i- Evalue: Independent E-value; K-turn: Kink-turn; Matt: Multiple Alignment with Translations and Twists; MD: Molecular dynamics; ML: Maximum Likelihood; Additional files MMC: Multi-member cluster; MSA: Multiple sequence alignment; ncRNA: Noncoding RNA; NR: Non-redundant; PDB: Protein Data Bank; Additional file 1: Table S1. RNA-binding proteins in 19 E. coli Pfam: Protein families database; RBD: RNA-binding domain; RBP: RNA-binding proteomes. All the RBPs obtained in the GWS of 19 E. coli strains have been protein; RNase PH: Ribonuclease PH; RNP: Ribonucleoprotein; listed in this table. The pathogenic and non-pathogenic E. coli strains RsmA: Repressor of secondary metabolites A; SCOP: Structural Classification have been highlighted in red and green, respectively. (DOC 115 kb) of Proteins; SMC: Single-member cluster; sRNA: Small RNA; TA: Toxin- Additional file 2: Table S2. Clusters of RNA-binding proteins obtained antitoxin; tRNA: Transfer RNA from 19 E. coli proteomes. The clusters of RBPs with more than one member in the GWS of 19 E. coli strains have been listed in this table. The RBPs were Acknowledgments clustered based on BLASTP searches at E-value, percentage identity and We thank NCBS (TIFR) for financial and infrastructural support. percentage query coverage cut-offs of 10−5, 30 and 70, respectively. (DOC 376 kb) Funding We thank University Grants Commission (UGC) and the NCBS Bridge Postdoctoral Additional file 3: RNA-binding proteins in all complete E. coli proteomes. Fellowship for funding P.G. All the RBPs obtained in the GWS of 166 E. coli strains have been listed here. The RefSeq IDs of the proteins are listed along with the total number of strains in which the protein is present mentioned in brackets. (DOC 185 kb) Availability of data and materials All the data related to this work, including accession IDs of proteins, have Additional file 4: 100 ns molecular dynamics simulations of the active been presented in the Additional files 1: Table S1, Additional file 2: Table S2 RNase PH monomer in the AMBER99SB protein, nucleic AMBER94 force and Additional file 3. field. The protein has been colour coded as in Fig. 5b. Hydrogen bonds at distance and angle cut-offs of 3 Å and 20°, respectively, have been Declarations shown at the region of interest with black dotted lines. (MP4 50,376 kb) All authors have gone through the manuscript and contents of this article Additional file 5: 100 ns molecular dynamics simulations of the inactive have not been published elsewhere. RNase PH monomer in the AMBER99SB protein, nucleic AMBER94 force field. The protein has been colour coded as in Fig. 5b. Hydrogen bonds Authors’ contributions at distance and angle cut-offs of 3 Å and 20°, respectively, have been RS conceived the idea and designed the project. PG acquired data and shown at the region of interest with black dotted lines. (MP4 67,789 kb) performed all the analyses. PG wrote the first draft of the manuscript and RS Additional file 6: 100 ns molecular dynamics simulations of the active improved on it. Both the authors read and approved the final version of the RNase PH dimer in the AMBER99SB protein, nucleic AMBER94 force field. manuscript. The protein has been colour coded as in Fig. 5b and c. Hydrogen bonds at distance and angle cut-offs of 3 Å and 20°, respectively, have been Ethics approval and consent to participate shown at the region of interest with black dotted lines. (MP4 67,624 kb) Not applicable, since this study has not directly used samples collected from humans, plant or animals, but has analysed publicly available, pre-existing Additional file 7: 100 ns molecular dynamics simulations of the inactive protein sequence data. RNase PH dimer in the AMBER99SB protein, nucleic AMBER94 force field. The protein has been colour coded as in Fig. 5b and c. Hydrogen bonds at distance and angle cut-offs of 3 Å and 20°, respectively, have been Consent for publication shown at the region of interest with black dotted lines. (MP4 67,820 kb) Not applicable. Additional file 8: 100 ns molecular dynamics simulations of the PELOTA_1 Competing interests domain from the ‘uncharacterised’ protein in complex with kink-turn RNA, in The authors declare that they have no competing interests. the AMBER99SB protein, nucleic AMBER94 force field. The protein has been represented in blue and the RNA in red. Hydrogen bonds at distance and angle cut-offs of 3 Å and 20°, respectively, have been shown between Publisher’sNote the protein and the RNA has been shown with black dotted lines. Springer Nature remains neutral with regard to jurisdictional claims in (MP4 54,830 kb) published maps and institutional affiliations. Additional file 9: 100 ns molecular dynamics simulations of the L7Ae K- turn binding domain from Archaeoglobus fulgidus in complex with kink- Received: 8 May 2017 Accepted: 9 August 2017 turn RNA from H. marismortui (PDB code: 4BW0: B), in the AMBER99SB protein, nucleic AMBER94 force field. The protein has been represented in blue and the RNA in red. Hydrogen bonds at distance and angle cut-offs of References 3 Å and 20°, respectively, have been shown between the protein and the 1. Kaper JB, Nataro JP, Mobley HLT. Pathogenic Escherichia Coli. Nat. Rev. RNA has been shown with black dotted lines. (MP4 66,564 kb) Microbiol. 2004;2:123–40. Additional file 10: Figure S1. Molecular phylogeny analysis of Cas6 2. Hacker J, Bender L, Ott M, Wingender J, Lund B, Marre R, et al. Deletions of proteins. a. All the proteins from Cluster 308 and Cas6 from E. coli strain chromosomal regions coding for fimbriae and hemolysins occur in vitro K12. b. Two previously uncharacterised proteins (UniProt IDs: C8U9I8 and and in vivo in various extra intestinal Escherichia Coli isolates. Microb Pathog. – C8TG04) from Cluster 308, with other known Cas6 proteins, including that 1990;8:213 25. from E. coli strain K12. In both the panels, the above-mentioned two 3. Hacker J, Kaper JB. Pathogenicity Islands and the evolution of microbes. – previously uncharacterised proteins from the pathogen-specific Cas6 Annu Rev Microbiol. 2000;54:641 79. proteins cluster (Cluster 308) have been highlighted in red and the 4. Hacker J, Blum-Oehler G, Muhldorfer I, Tschape H. Pathogenicity islands of Cas6 protein from E. coli strain K12 in blue. (JPEG 4531 kb) virulent bacteria: structure, function and impact on microbial evolution. Mol Microbiol. 1997;23:1089–97. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 22 of 23

5. Caprioli A, Morabito S, Brugère H, Oswald E. Enterohaemorrhagic Escherichia in virulence, antibiotic production and genomic flux in Serratia sp. ATCC Coli: emerging issues on virulence and modes of transmission. Vet Res. 39006. BMC Genomics. 2013;14:822. 2005;36:289–311. 37. Pessi G, Williams F, Hindle Z, Heurlier K, Holden MTG, Cámara M, et al. The 6. Garmendia J. Frankel gad CVF. Enteropathogenic and Enterohemorrhagic global posttranscriptional regulator RsmA modulates production of virulence Escherichia Coli infections. Infect Immun. 2005;73:2573–85. determinants and N-acylhomoserine lactones in Pseudomonas Aeruginosa. 7. Perez-Rueda E, Martinez-Nuñez MA. The repertoire of DNA-binding transcription J Bacteriol. 2001;183:6676–83. factors in prokaryotes: functional and evolutionary lessons. Sci Prog. 2012;95:315–29. 38. Liaw SJ, Lai HC, Ho SW, Luh KT, Wang WB. Role of RsmA in the regulation 8. Cusack S. RNA – protein complexes. Curr Opin Struct Biol. 1999;6:66–73. of swarming motility and virulence factor expression in Proteus Mirabilis. 9. Draper DE. Themes in RNA-protein recognition. J Mol Biol. 1999;293:255–70. J Med Microbiol. 2003;52:19–28. 10. Jones S, Daley DT, Luscombe NM, Berman HM, Thornton JM. Protein-RNA 39. Mulcahy H, O’Callaghan J, O’Grady EP, Maciá MD, Borrell N, Gómez C, et al. interactions: a structural analysis. Nucleic Acids Res. 2001;29:943–54. Pseudomonas Aeruginosa RsmA plays an important role during murine 11. Chen Y, Varani G. Protein families and RNA recognition. FEBS J. 2005;272:2088–97. infection by influencing colonization, virulence, persistence, and pulmonary 12. Hall KB. RNA – protein interactions. Curr Opin Struct Biol. 2002;12:283–8. inflammation. Infect Immun. 2008;76:632–8. 13. Schroeder R, Barta A, Semrad K. Strategies for RNA folding and assembly. 40. Mulcahy H, O’Callaghan J, O’Grady EP, Adams C, O’Gara F. The Nat Rev Mol Cell Biol. 2004;5:908–19. posttranscriptional regulator RsmA plays a role in the interaction between 14. Windbichler N, von Pelchrzim F, Mayer O, Csaszar E, Schroeder R. Isolation Pseudomonas Aeruginosa and human airway epithelial cells by positively of small RNA-binding proteins from E. coli : evidence for frequent interaction regulating the type III secretion system. Infect Immun. 2006;74:3012–5. of RNAs with RNA polymerase. RNA Biol. 2008;5:30–40. 41. Chao N-X, Wei K, Chen Q, Meng Q-L, Tang D-J, He Y-Q, et al. The rsmA -like 15. Aiba H. Mechanism of RNA silencing by Hfq-binding small RNAs. Curr Opin gene rsmA Xcc of Xanthomonas campestris pv. Campestris is involved in Microbiol. 2007;10:134–9. the control of various cellular processes, including pathogenesis. Mol. Plant- 16. De Lay N, Schu DJ, Gottesman S. Bacterial small RNA-based negative Microbe Interact. 2008;21:411–23. regulation: Hfq and its accomplices. J. Biol. Chem. 2013:7996–8003. 42. Vercruysse M, Köhrer C, Davies BW, Arnold MFF, Mekalanos JJ, RajBhandary 17. Gaballa A, Antelmann H, Aguilar C, Khakh SK, Song K-B, Smaldone GT, et al. UL, et al. The Highly Conserved Bacterial RNase YbeY Is Essential in Vibrio The Bacillus Subtilis Iron-sparing response is mediated by a fur-regulated cholerae, Playing a Critical Role in Virulence, Stress Regulation, and RNA small RNA and three small, basic proteins. Proc Natl Acad Sci U S A. 2008; Processing. Klose KE, editor. PLoS Pathog. 2014;10:e1004175. 105:11927–32. 43. Ghosh P, Sowdhamini R. Genome-wide survey of putative RNA-binding 18. Geissmann TA, Touati D. Hfq, a new chaperoning role: binding to proteins encoded in the human proteome. Mol BioSyst Royal Society of messenger RNA determines access for small RNA regulator. EMBO J. 2004; Chemistry. 2016;12:532–40. 23:396–405. 44. BatemanA,MartinMJ,O’Donovan C, Magrane M, Apweiler R, Alpi E, et 19. Holmqvist E, Vogel J. A small RNA serving both the Hfq and CsrA regulons. al. UniProt: a hub for protein information. Nucleic Acids Res. 2015;43: Genes Dev. 2013;27:1073–8. D204–12. 20. Van Assche E, Van Puyvelde S, Vanderleyden J, Steenackers HP. RNA-binding 45. Tatusova T, Ciufo S, Federhen S, Fedorov B, McVeigh R, O’Neill K, et al. proteins involved in post-transcriptional regulation in bacteria. Front. Update on RefSeq microbial genomes resources. Nucleic Acids Res. Microbiol. 2015;6. 2015;43:D599–605. 21. Liu JM, Camilli A. A broadening world of bacterial small RNAs. Curr Opin 46. Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ. Basic local alignment Microbiol. 2010:18–23. search tool. J Mol Biol. 1990;215:403–10. 22. Oliva G, Sahr T, Buchrieser C. Small RNAs, 5′ UTR elements and RNA-binding 47. Kaushik S, Mutt E, Chellappan A, Sankaran S, Srinivasan N, Sowdhamini R. proteins in intracellular bacteria: Impact on metabolism and virulence. FEMS Improved Detection of Remote Homologues Using Cascade PSI-BLAST: Microbiol Rev. 2015:331–49. Influence of Neighbouring Protein Families on Sequence Coverage. 23. Sauer E, Schmidt S, Weichenrieder O. Small RNA binding to the lateral Promponas VJ, editor. PLoS One. 2013;8:e56449. surface of Hfq hexamers and structural rearrangements upon mRNA target 48. Šali A, Blundell TL. Comparative protein Modelling by satisfaction of spatial recognition. Proc Natl Acad Sci. 2012;109:9396–401. restraints. J Mol Biol. 1993;234:779–815. 24. Prasanth KV, Spector DL. Eukaryotic regulatory RNAs: an answer to the 49. Laskowski RA, Macarthur MW, Moss DS, Thornton JM. PROCHECK: a program “genome complexity” conundrum. Genes Dev. 2007;21:11–42. to check the stereochemical quality of proteins structures. J Appl 25. Hannon GJ. RNA interference. Nature. 2002;418:244–51. Crystallogr. 1993;26:283–91. 26. Mattick JS. The Functional Genomics of Noncoding RNA. Science (80-. ). 50. Profiles T. VERIFY3D : assessment of protein models with three- dimensional 2005;309:1527–8. profiles. Methods Enzymol. 1997;277:396–404. 27. Sonnleitner E, Hagens S, Rosenau F, Wilhelm S, Habel A, Jäger KE, et al. 51. Wiederstein M, Sippl MJ. ProSA-web: interactive web service for the recognition Reduced virulence of a hfq mutant of Pseudomonas Aeruginosa O1. Microb of errors in three-dimensional structures of proteins. Nucleic Acids Res. Pathog. 2003;35:217–28. 2007;35:W407–10. 28. Sittka A, Pfeiffer V, Tedin K, Vogel J. The RNA chaperone Hfq is essential for 52. Pugalenthi G, Shameer K, Srinivasan N, Sowdhamini R. HARMONY: a server the virulence of salmonella typhimurium. Mol Microbiol. 2007;63:193–217. for the assessment of protein structures. Nucleic Acids Res. 2006;34:231–4. 29. Sharma AK, Payne SM. Induction of expression of hfq by DksA is essential 53. Wang J, Cieplak P, Kollman PA. How well does a restrained electrostatic for Shigella flexneri virulence. Mol Microbiol. 2006;62:469–79. potential (RESP) model perform in calculating conformational energies of 30. Ding Y, Davis BM, Waldor MK. Hfq is essential for Vibrio cholerae virulence organic and biological molecules? J Comput Chem. 2000;21:1049–74. and downregulates σE expression. Mol Microbiol. 2004;53:345–54. 54. Berendsen HJC, van der Spoel D, van Drunen R. GROMACS: a message- 31. Kendall MM, Gruber CC, Rasko DA, Hughes DT, Sperandio V. Hfq virulence passing parallel molecular dynamics implementation. Comput Phys regulation in enterohemorrhagic Escherichia Coli O157:H7 strain 86-24. Commun. 1995;91:43–56. J Bacteriol. 2011;193:6843–51. 55. Krissinel E, Henrick K. Inference of macromolecular assemblies from crystalline 32. Chao Y, Vogel J. The role of Hfq in bacterial pathogens. Curr. Opin. Microbiol. state. J Mol Biol. 2007;372:774–97. 2010. p. 24–33. 56. Dolinsky TJ, Nielsen JE, McCammon JA, Baker NA. PDB2PQR: an automated 33. Zeng Q, McNally RR, Sundin GW. Global small RNA chaperone Hfq and pipeline for the setup of Poisson-Boltzmann electrostatics calculations. regulatory small RNAs are important virulence regulators in erwinia Nucleic Acids Res. 2004;32:W665–7. amylovora. J Bacteriol. 2013;195:1706–17. 57. Unni S, Huang Y, Hanson RM, Tobias M, Krishnan S, Li WW, et al. Web servers 34. Christiansen JK, Larsen MH, Ingmer H, Sogaard-Andersen L, Kallipolitis BH. and services for electrostatics calculations with APBS and PDB2PQR. J Comput The RNA-binding protein Hfq of Listeria monocytogenes: role in stress Chem. 2011;32:1488–91. tolerance and virulence. J Bacteriol. 2004;186:3355–62. 58. Sowdhamini R, Sukhwal A. PPCheck: a Webserver for the quantitative 35. Geng J, Song Y, Yang L, Feng Y, Qiu Y, Li G, et al. Involvement of the post- analysis of protein–protein interfaces and prediction of residue transcriptional regulator Hfq in Yersinia pestis virulence. PLoS One. 2009;4. hotspots. Bioinform Biol Insights. 2015;9:141. 36. Wilf NM, Reid AJ, Ramsay JP, Williamson NR, Croucher NJ, Gatto L, et al. 59. Menke M, Berger B, Cowen L. Matt: local flexibility aids protein multiple RNA-seq reveals the RNA binding proteins, Hfq and RsmA, play various roles structure alignment. PLoS Comput Biol. 2008;4:e10. Ghosh and Sowdhamini BMC Genomics (2017) 18:658 Page 23 of 23

60. Dominguez C, Boelens R, Bonvin AMJJ. HADDOCK: a protein−protein docking 91. Deutscher MP, Marshall GT, Cudny H. RNase PH: an Escherichia Coli approach based on biochemical or biophysical information. J Am Chem Soc. phosphate-dependent nuclease distinct from polynucleotide phosphorylase. 2003;125:1731–7. Proc Natl Acad Sci. 1988;85:4710–4. 61. Edgar RC. MUSCLE: multiple sequence alignment with high accuracy and 92. Kelly KO, Deutscher MP. Characterization of Escherichia Coli RNase PH. J Biol high throughput. Nucleic Acids Res. 2004;32:1792–7. Chem. 1992;267:17153–8. 62. Kumar S, Stecher G, Tamura K. MEGA7: molecular evolutionary genetics 93. Wen T, Oussenko IA, Pellegrini O, Bechhofer DH, Condon C. Ribonuclease analysis version 7.0 for bigger datasets. Mol Biol Evol. 2016;33:1870–4. PH plays a major role in the exonucleolytic maturation of CCA-containing 63. Kumar S, Stecher G, Peterson D, Tamura K. MEGA-CC: computing core of tRNA precursors in Bacillus Subtilis. Nucleic Acids Res. 2005;33:3636–43. molecular evolutionary genetics analysis program for automated and 94. Jensen KF. The Escherichia Coli K-12 “wild types” W3110 and MG1655 have iterative data analysis. Bioinformatics. 2012;28:2685–6. an rph frameshift mutation that leads to pyrimidine starvation due to low 64. Jones DT. Protein secondary structure prediction based on position-specific pyrE expression levels. J Bacteriol. 1993;175:3401–7. scoring matrices. J Mol Biol. 1999;292:195–202. 95. Harlow LS, Kadziola A, Jensen KF, Larsen S. Crystal structure of the 65. Linder P, Jankowsky E. From unwinding to clamping — the DEAD box phosphorolytic exoribonuclease RNase PH from Bacillus Subtilis and RNA helicase family. Nat Rev Mol Cell Biol Nature Publishing Group. implications for its quaternary structure and tRNA binding. Protein Sci. 2011;12:505–16. 2004;13:668–77. 66. Blattner FR. The Complete Genome Sequence of Escherichia coli K-12. 96. Choi JM, Park EY, Kim JH, Chang SK, Cho Y. Probing the functional importance Science (80-. ). 1997;277:1453–62. of the Hexameric ring structure of RNase PH. J Biol Chem. 2004;279:755–64. 67. Kim S-H, Hung L-W, Wang IX, Nikaido K, Liu P-Q, Ames GF-L. No Title. Nature. 97. Anantharaman V. Comparative genomics and evolution of proteins involved 1998;396:703–7. in RNA metabolism. Nucleic Acids Res. 2002;30:1427–64. 68. Story RM, Steitz TA. Structure of the recA protein-ADP complex. Nature. 98. Anantharaman V, Iyer LM, Aravind L. Ter-dependent stress response systems: 1992;355:374–6. novel pathways related to metal sensing, production of a nucleoside-like 69. Abrahams JP, Leslie AGW, Lutter R, Walker JE. Structure at 2.8 Â resolution metabolite, and DNA-processing. Mol BioSyst. 2012;8:3142. of F1-ATPase from bovine heart mitochondria. Nature. 1994;370:621–8. 99. Huang L, Lilley DMJ. The molecular recognition of kink-turn structure by the 70. Dong J, Lai R, Jennings JL, Link AJ, Hinnebusch AG. The novel ATP-binding L7Ae class of proteins. RNA. 2013;19:1703–10. cassette protein ARB1 is a shuttling factor that stimulates 40S and 60S 100. Barrangou R, Marraffini LA. CRISPR-cas systems: Prokaryotes upgrade to ribosome biogenesis. Mol Cell Biol. 2005;25:9859–73. adaptive immunity. Mol Cell. 2014:234–44. 71. Samra N, Atir-Lande A, Pnueli L, Arava Y. The elongation factor eEF3 (Yef3) 101. Jiang F, Doudna JA. The structural biology of CRISPR-Cas systems. Curr Opin – interacts with mRNA in a translation independent manner. BMC Mol Biol. Struct Biol. 2015:100 11. 2015;16:17. 102. van der Oost J, Westra ER, Jackson RN, Wiedenheft B. Unravelling the 72. Rodnina MV. Protein synthesis meets ABC ATPases: new roles for Rli1/ structural and mechanistic basis of CRISPR-Cas systems. Nat Rev Microbiol. – ABCE1. EMBO Rep. 2010;11:143–4. 2014;12:479 92. 73. Van Melderen L, De Bast MS. Bacterial toxin-Antitoxin systems: More than selfish entities? PLoS Genet. 2009. 74. Van Melderen L. Toxin-antitoxin systems: Why so many, what for? Curr Opin Microbiol. 2010:781–5. 75. Goeders N, Van Melderen L. Toxin-antitoxin systems as multilevel interaction systems. Toxins (Basel). 2013:304–24. 76. Buts L, Lah J, Dao-Thi MH, Wyns L, Loris R. Toxin-antitoxin modules as bacterial metabolic stress managers. Trends Biochem Sci. 2005:672–9. 77. Gerdes K, Christensen SK, Løbner-Olesen A. Prokaryotic toxin-antitoxin stress response loci. Nat. Rev. Microbiol. 2005;3:371–82. 78. Jankowsky E, Fairman ME. RNA helicases–one fold for many functions. Curr Opin Struct Biol. 2007;17:316–24. 79. Jankowsky E. RNA helicases at work: Binding and rearranging. Trends Biochem Sci. 2011:19–29. 80. Hamma T, Ferré-D’Amaré AR. Pseudouridine Synthases. Chem Biol. 2006;13: 1125–35. 81. Phadtare S, Alsina J, Inouye M. Cold-shock response and cold-shock proteins. Curr Opin Microbiol. 1999:175–80. 82. Yamanaka K. Cold shock response in Escherichia Coli. J Mol Microbiol Biotechnol. 1999;1:193–202. 83. Fozo EM, Kawano M, Fontaine F, Kaya Y, Mendieta KS, Jones KL, et al. Repression of small toxic protein synthesis by the sib and OhsC small RNAs. Mol Microbiol. 2008;70:1076–93. 84. Louwen R, Staals RHJ, Endtz HP, van Baarlen P, van der Oost J. The role of CRISPR-Cas Systems in Virulence of pathogenic bacteria. Microbiol Mol Biol Rev. 2014;78:74–88. 85. Iyer LM, Koonin E V, Aravind L. No Title. Genome Biol. 2002;3:research0012.1. Submit your next manuscript to BioMed Central 86. Arthur DC, Ghetu AF, Gubbins MJ, Edwards RA, Frost LS, Glover JNM. FinO is and we will help you at every step: an RNA chaperone that facilitates sense-antisense RNA interactions. EMBO J. 2003;22:6346–55. • We accept pre-submission inquiries 87. Arthur DC, Edwards RA, Tsutakawa S, Tainer JA, Frost LS, Glover JNM. Mapping • Our selector tool helps you to find the most relevant journal interactions between the RNA chaperone FinO and its RNA targets. Nucleic • Acids Res. 2011;39:4450–63. We provide round the clock customer support 88. Ghetu AF, Gubbins MJ, Frost LS, Glover JN. Crystal structure of the bacterial • Convenient online submission conjugation repressor finO. Nat Struct Biol. 2000;7:565–9. • Thorough peer review 89. Mark Glover JN, Chaulk SG, Edwards RA, Arthur D, Lu J, Frost LS. The FinO • Inclusion in PubMed and all major indexing services family of bacterial RNA chaperones. Plasmid. 2015;78:79–87. 90. Iyer LM, Burroughs AM, Aravind L. The ASCH superfamily: novel domains • Maximum visibility for your research with a fold related to the PUA domain and a potential role in RNA metabolism. Bioinformatics. 2006;22:257–63. Submit your manuscript at www.biomedcentral.com/submit